[linux-cifs-client] CIFS mounted share: scandir returns incomplete list

[Sam Liddicott]

Interesting; having rolled back to an earlier samba4 there seem to be no
problems, and also wireshark is able to crack-open the smb packets ok.

I'll look further

Sam

* Sam Liddicott wrote, On 01/09/08 09:44:
> * Hervé Fache wrote, On 01/09/08 08:43:
>> In my case there are 677 files in the DOS file name format in that
>> directory...
> I think the problem shows up when more than once SMB packet is used to
> contain the responses, it may occur on a smaller number of files,
> depending on the negotiated max-xmit.
>>
>> Could you let me know how to log packets for this machine
>> ('abrolhos') and protocol? I could leave the backup server logging
>> all the time until the problem happens again and send you the
>> relevant log when it does then...
> something like: tcpdump -n -i eth0 -w dump.pcap -s 1500 port 445
>
> but if you can run wireshark directly on the machine, or a machine on
> that switch segment  you can use a capture filter like:
> host SERVER.IP and port 445
>
> Once you have a capture you can open it with wireshark and use a
> display filter:
> smb and (smb.trans2.cmd == 0x1 or smb.trans2.cmd == 0x2)
>
> and then export the filtered packets.
>
>>
>> Also, isn't there a way maybe to send the number of entries as read
>> on the CIFS server so we can at least detect the inconsistency? Or
>> does it break the protocol?
> The number of returned entries is part of the response in the first
> nttrans response packet.
> Somehow the re-combined nttrans response isn't correctly read;
> wireshark doesn't seem to manage to properly re-constitute nttrans
> responses either.
>
> Sam
>>
>> Hervé
>>
>> 2008/8/31 Sam Liddicott <sam at liddicott.com <mailto:sam at liddicott.com>>
>>
>>     I've noticed similar behavior on samba4 with vfs-cifs. By
>>     modifying the request to reduce the number of entries that
>>     find-first or find-next require I can make the problem "go away".
>>
>>     I think the problem is related to the encoding or decoding of
>>     multi-part nttrans requests.
>>
>>     I believe that Stefan Metzemacher is aware of alignment issues in
>>     the packing of large nttrans responses; your experience suggest a
>>     problem also in unpacking large response.
>>
>>     I've been to busy to follow this up, having to make do with the
>>     workaround I mention, but hope to spend time on it the latter
>>     part of this week.
>>
>>     However if you bring it up on irc when I'm online (samjam) and
>>     metze (I can't really speak for him) we may get somewhere. If you
>>     have some packet dumps that can be put through wireshark it may help.
>>
>>     The problem occurs for me when there are more than a thousand
>>     longish filenames in a directory.
>>
>>     sam
>>
>>     ------------------------------------------------------------------------
>>     From: Hervé Fache <Herve at lucidia.net <mailto:Herve at lucidia.net>>
>>     Sent: 31 August 2008 13:51
>>     To: linux-cifs-client at lists.samba.org
>>     <mailto:linux-cifs-client at lists.samba.org>
>>     Subject: [linux-cifs-client] CIFS mounted share: scandir returns
>>     incomplete list
>>
>>
>>     Hello there,
>>
>>     I have a backup system on my Linux server which mounts shares to
>>     back them up. To backup Windows shares, I use the CIFS mount type.
>>
>>     Every so often (hadn't happened in a few weeks until today),
>>     scandir returns an incomplete list of contents, without any error
>>     code being returned. Also, errno is not set. The problem happens
>>     on a machine connected via WiFi (Intel 2200BG).
>>
>>     If you have any clues or if I can turn on some debug to track the
>>     issue, please let me know.
>>
>>     Also, if you have ideas on how I could work around this issue,
>>     all ideas are welcome!
>>     Hervé
>>
>>
>

-------------- next part --------------
HTML attachment scrubbed and removed